3.e. Overview

Analysis categories include:

  1. Cross-Sectional Analyses – Identifying significant microbial differences between PL or PGH groups and saline controls at a given time point

  2. Longitudinal Analyses – Identifying significant microbial differences over time within each group

  3. Pre-Post Intervention Analyses – Identifying microbial changes in response to hormonal intervention (comparing pre- vs. post-intervention time points, incorporating multiple measurements per mouse)

  4. Microbiome-Metabolic Correlation Analyses – Assessing relationships between specific microbial taxa and various metabolic metrics:

Each section will include a brief methodological explanation before presenting the corresponding results.

3.e.i. Cross-Sectional Analyses

When comparing PGH or PL mice to Saline mice including all intervention day samples (D0 and beyond), we get no significant results.

3.e.ii. Longitudinal Analyses

Saline

I did not test for significant microbial differences between baseline and intervention days for mice given just saline, because I wasn’t sure what the value would be at this stage. But I am happy to if we think it would be beneficial!

PGH

I tested for significant longitudinal microbial differences in mice given PGH.

Phylum Level:

Here is the statistical info on the significant result above:

  • Increases were seen in the phyla Tenericutes and Bacteroidetes

  • Decreases were seen in the phylum Actinobacteria

Class Level:

At the class level, nothing popped up as significant after correcting for FDR, but it appears that enrichment in Tenericutes is driven by the class Mollicutes (unadjusted p = 0.022).

Order Level:

At the order level, nothing popped up as significant after correcting for FDR, but it appears that enrichment in Mollicutes is driven by an order called “RF39” (unadjusted p = 0.022).

Family AND Genus Level:

Note: Many of the family-level trends here could then be differentiated to the genus level. To make this document easier to follow, when applicable, I am going to nest those trends here, presenting the family information and then the related genus-level plots directly after.

  • There was a significant decrease in the family Streptomycetaceae, which is in the phylum Actinobacteria, which showed a general trend of decreased abundance.

This trend was predominantly driven by the genus Streptomyces.

  • Though directions of effect differed, significant effects were found in five different families within the order Clostridiales

    1. Lachnospiraceae (decrease)

(here’s another strong figure showing that Lachnospiraceae decrease, but from day 0 to 5):

This trend was driven by decreases in three different genera within this family:

  1. Epulopiscium

  1. Blautia

  1. Lachnospira

  1. Eubacteriaceae (increase)

This trend was predominantly driven by the genus Pseudoramibacter.

  1. Peptococcaceae (decrease)

This trend was predominantly driven by a genus noted as “rc4.4”.

  1. Dehalobacteriaceae (decrease)

(here’s another strong figure showing that Dehalobacteriaceae decrease, but from day 0 to 5):

This trend was predominantly driven by the genus Dehalobacterium.

  1. Clostridiaceae (increase)

(here is also a figure for Clostridiaceae increases in PGH mice from days 0 to 5):

This trend was predominantly driven by a genus noted as “SMB53”.

  • There was also a general decrease of Clostridiales that could not be differentiated to the family level

  • Additionally, the order “RF39” noted above, though not further differentiated, did pop up as significant at this level

Here is the statistical info on the significant result above:

Genus Level:

  • 8 of the genus-level analyses were included above under their corresponding family-level analyses that also were significant.

  • Excluding those, there were significant decreases in 4 genera:

    1. Faecalibacterium

  1. Jeotgalicoccus

  1. Lactococcus

  1. Oscillospira

  • There were also 4 new families that popped up as significant in this analysis that lack genus-level identification, despite not appearing in the family-level analysis round.
    • Of these, two are in the phylum Proteobacteria, which both increased:
    1. Rhodobacteraceae

  1. Methylobacteriaceae

  • The other two families are Firmicutes, and they decreased:
  1. Paenibacillaceae

  1. Ruminococcaceae (same family as the genus Faecalibacterium that also popped up separately above)

Here is the statistical info on the significant result above:

Species Level:

(Through D5): None of the groups identified above could be narrowed down by Maaslin beyond the genus level, so the species-level plots are all identical to those above.

(Through D5): We saw significant longitudinal decreases in the following species:

We saw significant longitudinal increases in the following species:

Here is the statistical info on the significant result above:

  • None of these could be identified at the species level, but both decreases came from the family Lachnospiraceae, with the first plot unable to differentiate beyond family and the second able to get down to the genus Lachnospira

  • Increases were seen in the order Clostridiales, with Maaslin also able to distinguish that some (but not all) of these increases were driven by members of the genus Ruminococcus.

  • Increases were also seen in the family Lactobacillaceae.

NOTE: The weirdness of our Day 8 samples could be driving these trends. ***

PL

I tested for significant longitudinal microbial differences in mice given PL. Because I was very skeptical of D8 values, I ran analysis at all levels that stopped at D5:

Phylum Level:

Here is the statistical info on the significant result above:

  • We see a mild longitudinal enrichment of Bacteroidetes and depletion of Firmicutes

  • We also see a mild enrichment of Verrucomicrobia

Class Level:

At the class level, nothing popped up as significant after correcting for FDR, but it appears that enrichment in Bacteroidetes is driven by the class Bacteroidia (unadjusted p = 0.027) and enrichment in Verrucomicrobia is driven by the class Verrucomicrobiae (unadjusted p = 0.044).

Order Level:

Here is the statistical info on the significant result above:

  • Depletion of Firmicutes appears to be dominantly driven by the order Lactobacillales

Family Level:

At the family level, nothing popped up as significant after correcting for FDR, but the depletion of Lactobacillales seems to be driven by the family Lactobacillaceae (unadjusted p = 0.00076).

Several other adjusted p values also were fairly low. See table below:

Genus Level:

Significant depletion was identified in the following genera of Firmicutes:

  • Lactobacillus

  • Catenibacterium

Additionally, the following Firmicutes families showed significant depletion, but only popped up as significant in the genus-level analysis, despite not being able to be identified beyond the family level:

  • Lactobacillaceae (as noted above)

  • Erysipelotrichaceae

Here is the statistical info on the significant result above (out of order from presentation of figures above):

Species Level (through D5):

None of the groups identified above could be narrowed down by Maaslin beyond the genus level, so the species-level plots are all identical to those above.

Species Level (through D8):

To show my suspicious of the D8 samples, this is an example of what one of the species-level trends look like including those days.

To me, this trend look clearly driven by D8 contamination.

3.e.iii. Pre-Post Intervention Analyses

Saline

I did not test for significant microbial differences between baseline and intervention days for mice given just saline, because I wasn’t sure what the value would be at this stage. But I am happy to if we think it would be beneficial!

PGH

I tested for significant microbial differences between baseline and intervention days for mice given PGH.

Phylum Level:

Here is the statistical info on the significant result above (it’s only one result so it’s a bit boring of a table):

PGH injected-mice had a significant enrichment in Tenericutes.

Class Level:

Here is the statistical info on the significant result above:

The enrichment in Tenericutes is predominantly within the Mollicutes class.

Order Level:

Here is the statistical info on the significant result above:

The enrichment in Mollicutes is predominantly within an order called “RF 39.”

Family Level:

Here is the statistical info on the significant results above:

  • Within that order “RF 39,” we are not able to get any additional information on the family, genus, or species that is enriched. Because of this I will not show the subsequent identical figures for the lower taxonomic levels.

  • There is also significant depletion of a family called Peptococcaceae

Genus Level:

Here is the statistical info on the significant results above:

Within Peptococcaceae, depletion is predominantly within a genus called “rc4.4”

Species Level:

Within that genus “rc4.4,” Maaslin cannot give species-level information, so the figures and stats are redundant and won’t be repeated.

PL

I tested for significant microbial differences between baseline and intervention days for mice given PL.

Phylum Level:

Here is the statistical info on the significant results above:

Matching the results we found in phyloseq, pl-exposed mice showed an enrichment in Bacteroidetes and a depletion in Firmicutes.

Class Level:

Here is the statistical info on the significant result above:

  • It appears the depletion in Firmicutes was spread across many lower level taxa, because no classes showed significance

  • The enrichment in Bacteroidetes was predominantly within the class Bacteroidia

Order Level:

Here is the statistical info on the significant result above:

The enrichment in Bacteroidia was predominantly within the order Bacteroidiales.

Family Level:

At the family level, Bacteroidiales enrichment appears to have been concentrated within two families, Porphyromonadaceae and one called “S24.7.” However, while unadjusted p values were significant (p = 0.005 and 0.009, respectively), with false discovery correction, neither reached significance.

Genus Level:

  • Within Porphyromonadaceae, the family that popped up as significant when p was unadjusted but not with false discovery correction, the majority of the effect came from the genus Parabacteroides (unadjusted p = 0.009)

  • From what I see online, Parabacteroides is considered a common gut commensal. It has also been found to “alleviate obesity and obesity-related dysfunctions in mice” (Wang et al, 2019 paper in Cell Reports), and has specifically been linked to glycemic control.

  • My quick lit search shows that Parabacteroides pops up a lot in GDM research, but is inconsistently found to be positively or negatively correlated. Kuang et al. (2017) did shotgun on healthy and GDM patients in the 2nd-3rd trimester (21–29 weeks) and found both that GDM patients had significantly higher Parabacteroides levels, but also that women with Parabacteroides abundance positively correlated with glucose levels during an OGTT. Their random forest model found that: “Bacterial species providing the highest discriminatory power were primarily members of the… Parabacteroides genera …consistent with our observation that Parabacteroides is the predominant genus accounting for differences in the gut microbiome between GDM patients and controls.” Dong et al. (2020) and Su et al. (2021) also found that GDM patients had a significantly higher relative abundance of Parabacteroides, as well as that “HOMA-IR increased with the higher abundance genus of Parabacteroides”. In contrast, Cortez et al. (2018) and Ma et al. (2020), found that healthy pregnant women had higher levels of Parabacteroides than GDM women. Ma et al. additionally found that women with higher Parabacteroides had lower fasting blood glucose levels.

  • In CD1 mice, Priyadarshini et al. (2017 – one of my fav papers) found that Parabacteroides were highest at D0 and gradually decreased during pregnancy, continuing to decrease through postpartum day 3.

Species Level:

Within Parabacteroides, the genus that popped up as significant when p was unadjusted but not with false discovery correction, the majority of the effect came from the species gordonii. Unadjusted p = 0.009).

3.e.iv. Microbiome-Metabolic Correlation Analyses

Glucose Metabolism

Endpoint OGTT AUCs

I had high hopes for this but nothing popped up as significant.

Endpoint ITT AUCs

The only significant effect found was at the level of order: Clostridia has a negative correlation with ITT AUC. A smaller ITT AUC indicates lower insulin resistance. So, mice with greater Clostridia levels also had greater insulin sensitivity. This effect held true when hormone both was and was not included as a random effect. The figure and stats below are for it not being included as a random effect.

Here is the statistical info for this result:

Endpoint fasting blood glucose levels

I ran these analyses two different ways: with hormone group included as a random effect, and not included.

When hormone is not included as a random effect, four taxonomic groups pop out as significant.

  1. Akkermansia muciniphila has a positive correlation with fasting BG levels.

  1. Plesiomonas shigelloides has a positive correlation with fasting BG levels. However, all but two samples had none of this bug, so the effect is driven by two mice with both high fasting BG and measurable levels of this bug.

  1. The family Coriobacteriaceae in the phylum Actinobacteria has a positive correlation with fasting BG levels.

  1. The genus Megamonas has a negative correlation with fasting BG levels. However, all but three samples had none of this bug, so the effect is driven by two mice with both low fasting BG and measurable levels of this bug.

Here is the statistical info for these results:

When hormone is included as a random effect, the significance for those four taxonomic groups remains fairly unchanged, with the exception of the p-value for Megamonas increasing from 0.02 to 0.002.

Body Composition

Lean mass

I was unsure the ideal way to run these analyses, so I did them a few ways. First, I included lean masses from both timepoints, with individual mouse and time point included as random effects. This is basically ignoring the effects of the hormones.

NOTE: These plots all say lean mass percentage as the x axis, but are actually showing absolute lean mass. I ran these analyses before I realized the EchoMRI returns absolute and not relative lean mass values.

I only ran this analysis as the genus level (cannot remember why), and found the following:

  • A negative correlation between Akkermansia (presumably muciniphila) and lean mass. However, this seems to be driven by two samples with very low lean mass and very high Akkermansia loads.

  • A positive correlation between the family Clostridiaceae and lean mass. This seems to be particularly driven by a genus called “SMB53.”

In previous mouse studies, including Zhang et al. (2024), “SMB53 showed a stronger positive correlation with body weight, white adipose tissues, liver weight and AA metabolites, and a simultaneously negative correlation with the anti-inflammatory cytokine IL-10.”

Here is the statistical info for these results:

I then analyzed lean mass only at endpoint, accounting for hormone group as a fixed effect. The output for this was a little funky – it didn’t make plots for me, but the full results table shows some interesting things:

Here are my takeaways:

  • SMB53, which was positively associated with lean mass, was (unsurprisingly) also positively associated with being a PGH mouse (p = 0.035).

  • 3 new genera popped up as having associations with lean mass:

  1. Prevotella – positively correlated, p = 0.019

  2. Jeotgalicoccus – positively correlated, p = 0.018 (was previously found to decrease in PGH, which is surprising)

  3. Coprococcus – negatively correlated, p = 0.005

  • I probably should go back and run this analysis at all taxonomic levels, but for now this is what I have.

Fat mass

Similarly, I started with looking at fat masses from both timepoints, with individual mouse and time point included as random effects. This is basically ignoring the effects of the hormones.

I found a positive correlation with the Proteobacterium Gallibacterium, but it was only at non-zero levels in 6 out of 36 samples.

When I analyzed fat mass only at endpoint, accounting for hormone group as a fixed effect, I found no significant associations.

Free and total water

I wasn’t super interested in these, but there were tons of significant correlations with both. I’ll include the significance tables for both at the genus level below:

Free water

And here are the figures (I’ll keep them tiny):

Total water

And here are the figures (I’ll keep them tiny):

Complete body composition heat map

I also looked at genus-level correlations between the four body comp variables and microbial taxa, and got this heat map:

I’m not sure how to interpret it though…

Weight

I started with looking at microbial associations with weight for ALL mice, not considering hormone or day as effects, but including mouse ID as a random effect. To me, this is asking if, ignoring the experiment itself, can variation in weight across our samples be correlated with certain microbial taxa?

I ran this at the phylum and genus level, and found many significant effects for both. I won’t include ALL of the figures here, but will show the full table of significant results, and highlight some key taxa.

Phylum Level:

These results seems to correlate with commonly understood microbial correlations with adiposity: higher Firmicutes, Proteobacteria, and Actinobacteria, and lower Bacteroidetes and Verrucomicrobia.

Here is the figure for Firmicutes:

Genus Level:

Many of these taxa also popped up on our longitudinal analyses of the hormone groups (e.g., Lactobacillaceae, Faecalibacterium, Clostridium, Megamonas, etc.)

Several are also well-studied in the context of metabolism and body mass, like Akkermansia, Blautia, Bacteroides, and Lactobacillus.

Here are the plots for Lactobacillus, Akkermansia, and Bacteroides:

Lactobacillus

Akkermansia

Bacteroides

I next wanted to run analyses looking to integrate this into our actual experimental design. We can’t really parse out cause and effect, but I wanted to know if either a) Hormone exposure-driven changes in weight produce corresponding shifts in microbiome composition, or b) Hormone exposure-driven changes in microbiome composition produce corresponding shifts in weight.

I wasn’t sure what to include as fixed and random effects, and I didn’t want to miss any interesting results, so I ran a bunch of different combinations (with mouse ID always included as a random effect):

  1. Hormone as a fixed effect with an interaction term (Weight:Hormone), and day as a random effect

  2. Analyses run on each hormone group independently with day as a random effect

  3. Analyses run on each hormone group independently with day as a fixed effect, filtering out the weird day 8 data

Here are the stats for analysis 1:

Phylum Level:

Genus Level:

The second analysis narrowed down our list of significant genera (see stats table below). Interestingly, one genus, “SMB53” within Clostridiaceae, which did not appear in our initial analysis list, now emerges as both positively associated with body weight AND with the PGH group. No other taxa appeared to be significantly associated with both weight and hormone group from this analysis.

Here are the associated figures for “SMB53”:

In the third analysis, day was included as a random effect and each hormone group was assessed independently.

In the PGH-dosed mice, significant weight-taxa correlations (done at genus and family level) were limited to “SMB53”, Akkermansia, Mogibacteriaceae, Bacillales, Streptophyta, and Lactobacillaceae.

In the PL-dosed mice, significant weight-taxa correlations (done at genus level) were primarily limited to members of Lactobacillales. Here is the full list:

As an example, here is the figure showing the correlation of Lactobacillus abundance and weight in PL mice:

This is relationship is also very clear at the order level of Lactobacillales:

In fourth analysis, day was included as fixed effect and day 8 samples were filtered out (for each hormone group independently).

In the PGH mice, “SMB53” appeared again, showing a significant increase over time and with increasing weight:

Other than that, the PGH correlations with time and with weight largely match those discussed above.

In PL mice, there was no overlap in the taxa that significantly changed over time and that significantly explained within-group weight variation. However, this analysis also highlighted the strong longitudinal decreases in the order Clostridiales in PL mice. I re-ran the analysis at the order level to make this figure demonstrating that:

Here’s the genus-level stats:

And here’s the order-level stats: